I've had some success in using Claude Code, with caveats.
To give some context - I started developing a tactical RPG. I had an MVP prior to using Claude Code. I continued to work on the project, but lost motivation due to work burnout and prioritizing other hobbies.
I gave Claude Code a try to see whether it's any use. It helped more than I expected it to - it helped me produce something while dealing with burnout by building on the MVP I developed prior to AI assisted development.
The main issues I ran into were:
1) A lot of effort into reviewing the output. Main difference from peer review is that there's quicker feedback.
2)It throws out some absolutely wild solutions sometimes. It build on my existing architecture, so it was easier to catch issues. If I hadn't developed the architecture without AI assistance, things could have gone badly.
3)I only pay for the $20 Claude plan. Anything useful Claude produces for me requires it to consume a lot of tokens due to back-and-forth questions and asking Claude to dig into source file.
The most significant issue I ran into with Claude is when it suggested solutions I don't have the background to review. I don't know much about optimization, so I ran into issues with both rendering and the ECS (entity component system) library. Claude gave me recommendations, but I didn't know how to evaluate the code due to lacking that experience.
Claude was good for things I know how to do but don't want to do. It's been helpful when I want to work on something without being motivated enough to put 100% (or even 70%) into it.
If it's things I don't know how to do (like game optimization) it's harmful.
I went to a DoD school from fourth grade until my second semester of 11th grade. After that, we moved to the U.S.
We moved to a good school district in the U.S, so the quality of the education remained the same. The most startling difference in a U.S public school was in how we were viewed by admin.
Compared to DoD schools, administrators in U.S public school system weren't too different from middle management at $corp. We were numbers on a spreadsheet.
A good analogy - U.S school admin acted like the kind of "manager" who judges you by the lines of code you produce and the number of commits you make. DOD school admin were the kind of people who judge you by the impact you made.
DoD schools respected our autonomy - we were treated like humans. Non-DoD schools treated us like cattle.
I'm a federal employee, working as a software engineer for the Department of Defense on embedded systems which are used on aircraft.
The first few months of DOGE were complete chaos. The senior executive service received conflicting information from one week to the next. Our operations were severely impacted without any benefit. We couldn't even go to test ranges for field testing.In addition to that, the five bullet points were a major security issue due to classification by aggregation.
Although DOGE is gone, we're still experiencing the fallout. There's more red tape than ever before. Everything requires multiple levels of approval - even ordering a replacement capacitor has to go up three levels of management. We're forced to bring defense contractors to our field tests because it's a fight to bring more than one federal employee, almost doubling the cost of any trip. It took half a year for us to even be allowed to mail equipment to various depots. Now, we're effectively forced to pay contractors for tasks we could do organically.
More privatization will drive up cost in the defense industry up significantly. I.E, an unnamed military contractor wants more than 5 million dollars for a line item breakdown for a quote they gave us.
It's not the DoD, but this is happening out in the open at DHS where the secretary is requiring her personal sign-off on any purchase over $100k[0] when the previous limit was $25M.
Deployments of critical resources, such as tactical and specialized search and rescue teams, were delayed as a result of a budget restriction requiring Homeland Security Secretary Kristi L. Noem to approve every purchase, contract and grant over $100,000[1]
> It's not the DoD, but this is happening out in the open at DHS where the secretary is requiring her personal sign-off on any purchase over $100k[0] when the previous limit was $25M.
This is EXACTLY what the PE firm did to my company after acquisition to "cut costs" and make numbers go up. I used to be able to sign off on my own purchase reqs up to $2000. That allowed me to easily acquire just about anything I needed and get my work done. Now I have to have EVERYTHING signed off on by upper management. As If I was an irresponsible spendthrift throwing money away on spare parts we actually needed. It's useless performative micromanagement by incompetent people. It's honestly insulting.
Government doesn't turn a profit, so the only way to get more money and prestige in the government is to have a higher headcount under you and more power. A good way to get more power is to add a bunch of red tape and also to slow down all the processes so you need more people and thus require a higher headcount which makes your management more prestigious.
All the incentives line up that this will only get worse.
"Unnamed military contractors" should be forced into competitive bids. Ditch the favoritism and let any US corporation bid on it and including bidding on the jobs where you need 3 layers of management to buy a capacitor when done by government employees. You can go out and buy a $500 Ar-15 on the private market that's about as good as an M4/M16 the military is paying thousands for -- that's the result of an actually free market for M-16/Ar-15 and the same thing can be brought to the rest of defense contracting.
Not defending DOGE, but it should be noted that much of what DoD pays for when it buys weapons is not the weapons, but reliable weapons and the logistics of supporting them under combat conditions. The apparently inflated unit costs often are given without looking at what's actually included contractually in those costs. Sure, any citizen can buy an AR-15 for $500, but that citizen isn't paying for a support structure that has to be able to deliver ammo or spare parts or spare units in a combat zone in time of war. Nor is that citizen paying for the guarantee that the contractor will be able to continue to produce all those things in the necessary quantities in time of war. Nor is the citizen paying for reliability guarantees about the weapon working under adverse conditions.
It's unfortunate that the reporting around such things doesn't actually dig into all this, but just gives quick sound bites without any real analysis.
I should have been more specific. I understand the sustainment cost of weapon systems and the like - which, while still too costly, is partially justified.
I was referring to the inflated cost the DoD pays for everyday items. I.E, having to pay double or triple the market rate for things like office chairs, computer headsets, and WIFI dongles. There's no sustainment cost. Just an inflated price.
I apologize in advance for not giving exact numbers - I'd rather not tie my posts to specific programs in the unlikely case that someone I work with reads my posts.
We received a quote containing two lines:
"Hardware Cost - xx million"
"Software Cost - xx million"
There was no further information. No detail on how those numbers were derived, nor what we were paying for other than the "hardware and software needed for organic sustainment of the system in question. The defense contractor wanted over 5 million for any additional detail. We wouldn't know whether we receive documentation, schematics, etc without paying additional money. I don't mean that they wanted us to pay more for things like documentation or schematics. We wouldn't know what we are receiving until we pay.
We were providing engineering support for a DoD Program Office which normally deals in multi-billion dollar acquisitions. They treated ~5 million as pocket change. Over multiple system acquisitions, that adds up fast.
To give some context - I started developing a tactical RPG. I had an MVP prior to using Claude Code. I continued to work on the project, but lost motivation due to work burnout and prioritizing other hobbies.
I gave Claude Code a try to see whether it's any use. It helped more than I expected it to - it helped me produce something while dealing with burnout by building on the MVP I developed prior to AI assisted development.
The main issues I ran into were:
1) A lot of effort into reviewing the output. Main difference from peer review is that there's quicker feedback.
2)It throws out some absolutely wild solutions sometimes. It build on my existing architecture, so it was easier to catch issues. If I hadn't developed the architecture without AI assistance, things could have gone badly.
3)I only pay for the $20 Claude plan. Anything useful Claude produces for me requires it to consume a lot of tokens due to back-and-forth questions and asking Claude to dig into source file.
The most significant issue I ran into with Claude is when it suggested solutions I don't have the background to review. I don't know much about optimization, so I ran into issues with both rendering and the ECS (entity component system) library. Claude gave me recommendations, but I didn't know how to evaluate the code due to lacking that experience.
Claude was good for things I know how to do but don't want to do. It's been helpful when I want to work on something without being motivated enough to put 100% (or even 70%) into it.
If it's things I don't know how to do (like game optimization) it's harmful.